Decoding indoor radon: An explainable AI approach to quantifying building, environmental, and inhabitants'...

Indoor radon exposure represents a significant environmental health hazard and is a recognised leading cause of lung cancer, characterized by complex, multifactorial origins. Despite its importance, a substantial portion of the variability in indoor radon concentration (IRC) remains unexplained by traditional models. This study, for the first time, leverages the combined power of explainable AI (XAI) and a comprehensive, high-quality dataset from the Second Italian National Radon Survey to address this critical gap. Utilizing the XGBoost algorithm, we achieve a level of IRC prediction performance that significantly exceeds previous efforts. Crucially, by analyzing the predictive model with the SHAP (SHapley Additive exPlanations) algorithm, we quantify the relative contributions to IRC variability of building characteristics, environmental factors, and, most notably, inhabitants' characteristics - an aspect largely neglected in the existing literature. Robustness of the approach was confirmed by a sensitivity analysis and a comparison with a different, largely linear, model (Lin-NN). Even with this advanced workflow, a notable amount of variance remains unexplained, highlighting the inherent limitations in predicting IRC. This residual variance may be attributable to the influence of inhabitant behavior and other highly localized factors specific to individual dwellings. This underscores the need for continued research into specific behavioral and other unobserved factors. Importantly, the findings are likely generalizable to other countries and contexts owing to Italy's diversity in environmental, building, and human-related factors, and the use of a multi-model approach. The findings may provide practical guidance to mitigate radon risk, by identifying building types associated with higher radon concentration and quantifying location-based risk, thereby supporting more effective, targeted mitigation strategies and potential building code updates, including mandatory radon-preventive measures for specific building categories.